Visual Analytics is a field of study that seeks to aid human cognition in the process of analyzing data. It aims at doing so through visual interfaces and automated computations. Such automated computations often translate to Machine Learning algorithms. Users can leverage from these algorithms in order to find interesting patterns in massive and unstructured data. These algorithms are, however, still often regarded as black-boxes i.e. mathematical models which are hard to interpret and difficult to interact with. In a single Machine Learning library more than 40 variants of algorithms can be found, and some with up to 14 different parameters. The Visual Analytics community has pointed out a lack of research in helping users set parameters and compare results between algorithms. Moreover, it is argued that most technology in the field, which has aimed at bringing transparency to black-boxes, is embedded in domain-specific systems thus making it hard to reach and use for other users and in other domains. In general, Machine Learning has an accessibility challenge: it is hard to use, to interpret and, if not either of these, to reach. This research proposal motivates the need for research in making Machine Learning more accessible for exploratory data analysis, reviews existing work in the field, presents a research plan and, finally, describes preliminary results.
Research proposal, PhD programme, University of Skövde